An M/M/1 queue has a Poisson process of arrivals of rate , one server, independent exponential service times of rate , and unlimited waiting room. Arrivals are independent of service times. The two M symbols denote the memoryless arrival and service assumptions; the final one denotes the number of servers. Under a work-conserving service discipline, the queue population , including the customer in service, is a birth-death process with ratesThe server has no departure clock when the queue is empty. The traffic intensity is the server utilization in equilibrium when . Detailed balance then gives the Stationary distribution , whose mean is . For example, three arrivals per minute and four services per minute give and mean queue population three. Raising arrivals to per minute raises that mean to nineteen: operation near capacity amplifies congestion. These formulas count the customer being served as well as those waiting.[ref]
https://ocw.mit.edu/courses/2-854-introduction-to-manufacturing-systems-fall-2016/927056a1af54772a587fd84ad4951e71_MIT2_854F16_Mm1Queue.pdf. MIT 2.854 M/M/1 Queue notes, pages 1-2, system population and waiting population
With packet arrival rate lambda, independent exponential payloads of mean one and byte-service rate C, let . The number N of unfinished packets has law . Conditional on N=n, the remaining payloads are n independent unit-mean exponentials, including the in-service residual by memorylessness. Summing their Laplace transforms gives . Thus W has mass at zero and, conditional on positivity, is exponential with rate . The probability of at least r headers is instead for integer r.
If potential arrivals form a Poisson process of rate and an arrival seeing customers joins with probability , the queue length is a birth-death process with birth rate and death rate for . Its stationary ratios, when normalizable, are
In a stationary M/M/1 queue with independent Poisson arrivals of rate and exponential service times of rate , the departure counting process is a Poisson process of rate . Moreover, the departures before a fixed time are independent of the queue population . This independence concerns the past departures: future departures depend on whether the queue is currently empty.
The proof uses time reversal. The stationary Geometric distribution satisfies Detailed balance, , so the stationary queue population is a Reversible Markov chain. Departures in forward time are arrivals in reversed time. In the reversed chain the upward rate is again the constant , so those arrivals are an independent Poisson process, including independence of their future increments from the present queue population. Reversing time gives both assertions. Consequently, feeding the output of a stationary M/M/1 queue into another queue supplies Poisson input to that second queue. The claim depends on equilibrium; a queue started empty has no immediate departures even though arrivals have positive rate.
In an M-M-1 queue with Bernoulli feedback, external arrivals form a Poisson process of rate , service times are independent exponential random variables of rate , and each completed service independently returns its customer with probability . A returned customer immediately rejoins the same queue and requires a fresh service time. The queue population does not change at a feedback completion. Only completions that leave permanently produce downward jumps, at rate while the queue is nonempty.
For a continuously busy server, independent Bernoulli thinning of its Poisson completion clock gives an exponential time of rate until the next permanent departure. Thus the queue population has exactly the same birth-death process as an M/M/1 queue with service rate . It has a Stationary distribution precisely when ; its geometric parameter is then . This reduction describes the population and external departures, while individual customers can undergo several visits. For example, doubles the mean number of services needed per external customer from one to two.
In equilibrium, a Bernoulli-feedback queue has the same queue-length process as an ordinary queue with service rate , so its external departures form a Poisson process of rate .
The busy period of an M/M/1 queue starts with an arrival into an empty queue and ends the next time the queue becomes empty. Assume independent Poisson arrivals of rate and exponential service times of rate . If , its mean isOne derivation follows the family of customers generated during services. The initial customer's mean service time is ; during that service, the expected number of arrivals is . Each new arrival generates further arrivals during its own service. The expected total service time is therefore the nonnegative geometric series , which gives the formula. This argument can first be applied to finitely many generations and then justified by monotone convergence.
When , the busy period is finite almost surely but has infinite mean. When , it has probability of never ending; its extended mean is also infinite. Thus the divergence at critical load reflects a change in return behavior, not an infinite mean individual service time.
Articles by others on the same topic
There are currently no matching articles.